Improved Walrus Optimizer for Feature ModeDecomposition Parameter Selection in BearingFault Feature Extraction

Abstract

Feature Mode Decomposition (FMD) is sensitive to parameter settings, which are often selected empirically and may limit bearing fault feature extraction. This study proposes an Improved Walrus Optimizer-assisted FMD (IWO-FMD) framework for bearing fault feature enhancement. The framework is designed for approximately fixed-speed conditions with known fault characteristic frequencies. IWO-FMD jointly optimizes FilterSize, CutNum, ModeNum, and MaxIterNum. Parameter search and held-out evaluation are performed on separate, non-overlapping vibration segments. Experiments on 12 kHz drive-end signals from the Case Western Reserve University (CWRU) bearing dataset cover 20 load–fault conditions with ten repeated runs per condition. IWO-FMD is compared with fixed-parameter FMD and five optimizer-assisted FMD methods. All optimizer-assisted methods use identical parameter ranges and 100 true objective-function evaluations. Compared with fixed-parameter FMD, IWO-FMD achieves higher mean fault characteristic peak (FCP) and signal-to-noise ratio (SNR) values in all 20 conditions and higher mean correlated kurtosis (CK) values in 19 conditions. After Benjamini–Hochberg correction, the SNR and CK differences are statistically significant in 19 and 17 conditions, respectively. The pooled fault-detection rate increases from 45.0% to 60.5%, while the candidate-level false-alarm rate decreases from 10.8% to 7.5%. Overall, IWO-FMD improves FMD parameter selection and provides a better balance between bearing fault feature enhancement and false-alarm control.